October 24, 2001

Future Tense

The first step in improving profitability is to understand the costs and benefits of current business decisions. Data mining can help.

By J.T. Lehman

Some of your competitors are using data mining right now. You may be using it, too. But if you're still not clear about how to use data mining to the best of your business advantage, read on.

In this article, I'll explain the connections between business issues and data mining by analyzing the economic consequences of a business strategy - using the specific example of claims investigation - enabled by data mining. As you'll see, data mining, when executed correctly, can improve the efficiency and profitability of a business process by helping you understand the true costs and benefits of your business decisions. (This article is designed for technically informed business managers, not data analysts.) In essence, using data mining to predict key future events gives you a clearer picture of the costs and benefits of a business strategy. From there, setting business strategy can be as easy as doing what will probably make money, and not doing what will probably lose money.

In the example of claims investigation, business-driven data mining makes it possible to increase profits by estimating how much cost an investigation would save. There are two methods here: First, for the transactions that are not scheduled for an investigation, you can make money by performing the investigations that save claim cost. Second, for the transactions that are scheduled for an investigation, you can make money by stopping the investigations that would not save claim cost.

In the insurance industry, transactions are often investigated to determine if a claim is valid. Investigations of claims also occur in other industries. For example, the healthcare industry investigates medical claims, the credit card industry investigates charge-backs, the automotive industry investigates warranty claims, and retailers investigate sales transactions. Furthermore, within these industries there are many types of investigations, each offering a chance for economic improvement. The insurance industry alone is typified by several investigation types, including independent medical examinations (IMEs), special investigations, surveillance operations, and medical record audits.

Each type of investigation involves a different level of effort, and often, the different types of investigations reveal entirely different types of information. However, the framework I'll describe here applies to each of these types of claim investigations in each of these industries, as long as they involve material costs as well as a sufficient amount of data to support data mining (as most of them do).

BUSINESS DECISION COSTS AND BENEFITS

In order meet its goal of improving business process return on investment (ROI), data mining must result in some change in business decision-making. Therefore, the first step in creating a general framework for business-driven data mining is to understand the economic impact of specific business decisions. We'll start with the costs and benefits that can be determined with existing data.

Claim investigations cost money. These costs are incurred from sources inside and outside the company servicing the claim. Money paid to service providers outside the company servicing the claim are called "hard costs" because they are explicitly represented in bills and contracts. Outside service providers are used both to provide independent assessments of claims, as in the case of an IME, and to provide offsite services for existing personnel, as in the case of surveillance far from the home office. Costs incurred from inside sources are called "soft costs" because they are often bundled into overhead.

Hard costs paid to outside service providers usually exceed soft costs of internal activities. For example, some bodily injury insurance policies grant the insurer the privilege to request that a claimant appear before an independent medical provider for a medical examination. A recent study found that the external costs of an IME in a given state averaged $350. Internal costs were assumed to be nearly half that - $150 - making the total cost of an IME about $500. In this study, the vast majority of IMEs had the same cost, but a small percentage of claim investigations had a lower cost because the claimant did not appear for the claim investigation. Because the overwhelming majority of claim investigations had the same cost, for the purposes of business strategy we will consider the cost of all claims to be the same.

There are other less obvious costs involved. Investigations make claimants unhappy; if the unhappy claimants are customers, they may terminate their relationship with the insurer. Figure 1 breaks down customers according to claims and investigations. Analysts can estimate the average length of the customer relationship for customers who:

1. Never file a claim

2. File a claim that is not investigated

3. File a claim that is investigated that results in a good outcome

4. File a claim that is investigated that results in a bad outcome.

The difference between 1 and 2 is the cost or benefit of the claims payment system apart from any issues of investigating claims. This key performance indicator reveals the hard dollar consequences of existing processes to pay claims. Because this article covers the decision whether or not to investigate once a claim has been filed, the main point compares 2 to 3 and 2 to 4.

The difference between 2 and 3 is the effect of a "good" decision to investigate. Customers whose claims are investigated and reduced in cost probably feel outraged. Of course, the difference in claim cost probably offsets the loss in customer value. The difference between 2 and 4 is the effect, presumably the cost, of a "bad" decision to investigate. Customers whose claims are investigated with no change in the claim probably feel that the company is trying to avoid its obligation, and some will take their business elsewhere.

Next, I'll outline the business decisions involved here and the known costs and benefits of those decisions. This information is necessary to determine the economic consequences of changing a business decision. In the following section, I'll describe the costs and benefits that are not yet known.

THE ROLE OF DATA MINING

We can deconstruct the expected value of an investigation into the probability of a good outcome multiplied by the value of a good outcome, plus the probability of a bad outcome multiplied by the value of a bad investigation minus the cost of the investigation. For example, if you bet on black at the roulette table, ignoring the green 00 slot, you have a 50 percent chance of winning double your bet and a 50 percent chance of losing your bet. The expected value is therefore 50% x 2 + 50% x 0 = 1, the amount of your bet.

Thus, here's what we know at this point: the value of a good outcome, and the value of a bad outcome. Since the probability of a bad outcome is one minus the probability of a good outcome, the only thing left to discover is the probability of a good outcome. That's where predictive data mining models come in.

In essence, a predictive model is a bridge between what is known and what is not known. Predictive models forecast a future event as a function of what is known now. Predictive models are built using past examples of the future event. Weather forecasting is a well-known example. Meteorologists use computer programs to predict future weather based on observations of current weather and the knowledge of past weather patterns.

In business, a predictive model is first built by data miners, and then deployed to business users via business intelligence software. During construction, data miners determine the relationship between what is known prior to the event of interest and the event of interest itself. In deployment, a predictive model estimates the event of interest; in this case, the probability that a claim investigation will have a good outcome.

The most effective predictive models are built on large amounts of reliably recorded, clean data. Required data includes the presence or absence of a claim investigation and the outcome of the claim investigation. Everything that was known prior to the claim investigation improves the model.

• Privacy
• Your California Privacy Rights
• Copyright © 2008 United Business Media LLC